Input data consists of a set of labelled JPG images of turtle faces from three different perspectives: top, left, right. There are 100 distinct turtle identities present in the training set. Some additional details about the training set are available in the tutorial Colab notebook.
An additional set of labelled images has also been provided. This set does not contain information about the position of the turtle’s face captured, but you may still find it useful.Some test images correspond to turtles in the training dataset and others do not.
For a given query image, return either the identity of the turtle in the image (as specified in the training data labels) or return the ‘new_turtle’ label.
A tutorial notebook is provided, this notebook shows you where to download the data, how to create a model and outputs a submission file which you can submit to this competition.
Here is the tutorial notebook.
We recommend you upload the notebook to Colab. The data is stored in a Google Bucket so by uploading the notebook to Colab you will not need to download the data.
If you would like to download the data straight to your local machine you can use these links, however, it is recommended you use the tutorial notebook as it doesn’t require you to download the data.
Disclaimer: Images can contain pictures of injured turtles.
Many of the turtles in this project are turtles who have been caught as bycatch by fishermen and bought to Local Ocean Conservation for rehabilitation or have been rescued by a LOC staff member. Each rescued turtle is assessed, then measured, weighed and tagged. If it is in good health, the turtle is transported to the Watamu Marine National Park where it is released back into the ocean. Severely injured turtles are admitted to the rehabilitation unit at LOC.
The data was collected through the Watamu Turtle Watch and Local Ocean Conservation.